Benchmarks
357 sources · 1,263 models × 3,079 benchmarks · 162,127 scoresPartial-credit companion to FrontierCUA v1 pass@1 on the same 40 enterprise computer-use workflows: the mean over workflows of each workflow's graded attempt score, the weighted share of rubric checks met on a 0-1 scale. The leaderboard uses it to break pass@1 ties.
267th strongest of 818 fitted benchmarks
Signal-to-noise +8.69 for the FrontierCUA v1 pass@1 family
Another benchmark represents this family in the fit.
| # | Model | Setting | Score | Source |
|---|---|---|---|---|
| 1 | Gemini 4 Carbon | high | 0.6 | |
| 2 | GPT-6.2 Astra | high · Codex | 0.6 | |
| 3 | GPT-6.1 Sol | high · Codex | 0.5 | |
| 4 | GPT-6 Astra | high · Codex | 0.5 | |
| 5 | GPT-6.2 Sol | high · Codex | 0.5 | |
| 6 | Claude Fable 5.1 | high | 0.5 | |
| 7 | Claude Opus 5.5 | high | 0.4 | |
| 8 | Claude Fable 5 | high | 0.4 | |
| 9 | Claude Opus 5 | high | 0.4 | |
| 10 | MiniMax M3.1 Flash | max | 0.4 | |
| 11 | Muse Wonder | high | 0.4 | |
| 12 | Muse Ember | high | 0.4 | |
| 13 | Qwen4-Max | high · Qwen-Agent computer-use harness (OSWorld-V2 QwenAgent port) via OpenRouter | 0.4 | |
| 14 | Claude Fable 5.2 | high | 0.4 | |
| 15 | Kimi K3 | high | 0.4 | |
| 16 | Sonnet 5 | high | 0.4 | |
| 17 | MiMo-V2.6-Flash | high | 0.3 | |
| 18 | GPT-6 Sol | high · Codex | 0.3 | |
| 19 | GPT-6.1 Terra | high · Codex | 0.3 | |
| 20 | Kimi K3.1 Pro | high | 0.3 | |
| 21 | MiMo-V2.6-Pro | high | 0.3 | |
| 22 | Grok 4.7 | high | 0.3 | |
| 23 | GPT-5.6 Sol | high · Codex | 0.3 | |
| 24 | GLM-5.3 Flash | high · GLM-5.3-FlashX inference-speed tier via Alibaba Cloud | 0.3 | |
| 25 | Muse Spark 1.3 | high | 0.1 |
Benchmarks outside its family whose model rankings track it most reliably (Spearman ρ over at least 10 shared models).